Refined Contact Map Prediction of Peptides Based on GCN and ResNet

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Peptide inter-residue contact map prediction is vital for determining peptide structure topology.
• Existing methods struggle with accuracy, particularly for long-range residue distances, due to limited homologous structures.

Purpose of the Study:

• To develop a novel deep neural network framework to refine inaccurate peptide contact maps.
• To improve the prediction accuracy of long-range residue-residue contacts in peptides.

Main Methods:

• A deep neural network framework was developed to refine existing contact maps.
• A residue graph was constructed from rough contact maps and processed using a graph convolutional neural network (GCN) to capture global information and long-range contacts.
• A residual convolutional neural network was incorporated for learning local information.

Main Results:

• The proposed framework successfully refines rough contact maps generated by existing methods.
• Experiments on four datasets showed significant improvements in predicting inter-residue long-range contacts.
• The method effectively captures global and local residue relationships for accurate contact map prediction.

Conclusions:

• The novel deep learning framework demonstrates effectiveness in enhancing peptide inter-residue contact map prediction.
• The integration of GCN and residual CNNs improves the capture of both long-range and local residue interactions.
• This approach offers a promising advancement for computational biology and peptide structure determination.